Background
The global demand for dairy products has increased rapidly, leading to high-density cattle farming where dystocia is a major concern for both animal welfare and farm profitability. Traditional sensor-based prediction methods (e.g., tail sensors, temperature probes) help but often cause discomfort, fail frequently, or are too costly. Recent studies have turned to computer vision to detect prepartum behaviors, but small-scale signs like tail raising or head-to-abdomen movements remain underexplored due to detection challenges.
Aims
This study proposes a deep learning-based computer vision approach for calving prediction, focusing on key behaviors: Tail Raising, Head to Abdomen, and Lying-Standing Transition. By using DeepLabCut for body-point tracking and behavior frequency analysis, the goal is to improve early dystocia detection in practical farm settings, reducing risks to both calves and cows.